☀ TL;DR

Trump signs order creating a Super Intelligence Force

Trump signed an executive order creating a "Super Intelligence Force" chaired by Jay Clayton, requiring a 120-day report on superintelligence risks and opportunities.

Bloomberg Intelligence says gains by DeepSeek and other Chinese labs have narrowed the US AI performance lead to a record low.

Proofpoint says TA419, a suspected China-aligned group, impersonated an Anthropic researcher to phish US AI policy experts with malware.

A White House AI task force will report on AI technology risks and federal government responsibility.

SoftBank CEO Masayoshi Son issued a rare warning about AI safety risks as machines gain abilities.

Lead · Industry

Trump Signs Order Creating Super Intelligence Force

Trump signed a Sept. 29 executive order creating a "Super Intelligence Force" chaired by DNI Jay Clayton, tasked with a 120-day report on superintelligence risks and opportunities, and directing the term "AI" be replaced by "SI."

Details →

Industry

24
24h Key Points
  1. TA419 · is a suspected China-aligned group that, per Proofpoint, impersonated an Anthropic researcher and former US officials to phish AI policy experts at think tanks, universities and law firms for cloud accounts.

  2. Homa · is a new transport protocol for AI clusters aiming to replace TCP and improve network performance for large-scale distributed training, with papers published at venues such as USENIX ATC.

  3. Micron · its CEO says memory supply will be much tighter in 2027 and 2028 than in 2026, signaling a worsening shortage driven by AI demand.

  4. Google · launched its first experimental satellite under Project Suncatcher, carrying four TPU chips, to test the feasibility of AI data centers in orbit.

  5. Google · researchers' RRSI method curbs self-improving agents' memorization of test tasks, lifting unseen-benchmark scores by up to 4.7 points while using about 30% fewer tokens than an unregularized version.

  6. Trump · signs an executive order creating a "Super Intelligence Force" chaired by DNI Jay Clayton, which must report within 120 days on superintelligence risks and opportunities and replace the term "AI" with "SI".

  7. South Korean banks · suffered intrusions, with about 25,000 people's credit data leaked at Shinhan Bank plus 119 KB Kookmin and 89 Hana customers; experts suspect AI automation enabled broad vulnerability scanning.

  8. Kyojin · is an ExLlamaV3-based engine for AMD Strix Halo (ROCm) that runs two 300B-class MoE models each on a single 128GB mini PC, with GLM-5.3-Flash prefilling at ~580 tok/s.

View all →

Voices

X · 24
24h Key Points
  1. Demis Hassabis · says he is proud of his team's work using AI to accelerate science and medicine for society.

  2. Guillermo Rauch · argues human-facing documents like READMEs should be written by hand, while agent-facing internals can be in AI English.

  3. Aravind Srinivas · says custom vertical AI apps can be built inside Computer, such as guessing an image's location with 3D and satellite views.

  4. Alexandr Wang · recalls teaching HRT executives Snapchat as an intern a decade ago, noting their quarterly revenue now exceeds Snap's market cap, and advises interns to teach them muse.

  5. hardmaru · notes Schmidhuber's formal theory of fun and creativity, linking compression progress to beauty, curiosity, art and jokes, appeared in a Japanese journal in 2009.

  6. Aravind Srinivas · says a run cleared Pokémon FireRed's Elite Four and Champion in one shot using Perplexity's Decisions API, with 592ms median response and $0.028 estimated inference cost.

  7. Bindu Reddy · says OpenAI confirmed the Astra 6.1 launch next week, calling it a good time to cut prices, as models self-improve and become more efficient.

  8. Bindu Reddy · argues open source models have an 8-12 week window to improve into a self-improving loop or be left behind, with frontier labs already in that loop.

  9. Teknium 🪽 · responds to the Hermes Gadget open SDK launch, a small device that talks to your own Hermes, with a desktop simulator running the same code as the ESP32 firmware.

  10. Yuchen Jin · argues agents changed how terminals are used, since you state intent and the agent operates the machine, while systems knowledge remains a moat.

  11. hardmaru · observes the AI consciousness debate reached the Vatican, with the West asking if Claude has a soul and the East asking if the tool works, explaining differing AI vibes.

  12. Simon Willison · argues in a new post that default hard budget caps are needed on pretty much everything.

  13. Guillermo Rauch · argues security is becoming a larger function in software companies, both verification engineering and capital allocation, a challenge and opportunity for startups.

  14. StepFun · announces Step 5 Preview landed on the Vals leaderboard, ranking 7th among open-weight models at $2.54 per task.

  15. Amjad Masad · describes models training smaller domain-specific replacements on the fly, likening it to a just-in-time compiler.

  16. Ethan Mollick · observes that US labs accuse Chinese labs of distillation, while Europe's sovereign model is fine-tuned on data generated by GLM and Qwen.

  17. yetone · announces magpie's plugin now supports both OpenCode plugins and pi extension packages, so one npm package handles login and requests across agents.

View all →

Launches

PH · HN · 12

CoreSpeed

One MCP endpoint giving agents apps, memory, and tools.

▲ 144 · Product Hunt · 24 comments ·

Blume 2.0

Open-source docs framework for humans and agents, zero-config production-grade docs sites

▲ 125 · Product Hunt · 2 comments ·

Gemini 4 Argon

Google's frontier model for long-horizon professional reasoning work.

▲ 96 · Product Hunt · 3 comments ·

Sellio

AI customer support in a shared inbox across chat, WhatsApp, and email.

▲ 82 · Product Hunt · 7 comments ·

Octri.dev

Upload an OpenAPI spec to generate docs, SDKs in ten languages, and an MCP server

▲ 81 · Product Hunt · 16 comments ·

Thinking Orbs

React component library of animated AI agent status orbs for developers.

▲ 81 · Product Hunt · 3 comments ·

ChatGPT Space

Shared canvas to create and plan together with ChatGPT.

▲ 80 · Product Hunt · 2 comments ·

eu/jev

The first Jev-like model hosted in the EU, GDPR compliant and not saving prompts.

▲ 83 · Product Hunt · 3 comments ·

Notchware

Turns the MacBook notch into a Dynamic Island for music controls and AI agent alerts.

▲ 80 · Product Hunt · 1 comments ·

Open Source

GitHub Trending · 20

Marketing skills for Claude Code and AI agents covering CRO, copywriting, SEO, and analytics.

GitHub Trending ·
★ 52.8k+345 today

World's first open-source agentic video production system with 12 pipelines and 100+ tools.

GitHub Trending ·
★ 62.8k+292 today

Claude Code, Codex and other harness ports of Poteto's pstack agent workflows.

GitHub Trending ·
★ 1.1k+242 today

Local inference engine for DeepSeek 4 Flash and PRO on Metal, CUDA and ROCm.

GitHub Trending ·
★ 23.3k+211 today

Garry Tan's Claude Code setup with 23 tools acting as CEO, Designer, Eng Manager and QA.

GitHub Trending ·
★ 135.0k+121 today

Agent skills library for generating, inspecting, and slicing CAD and robot-description artifacts.

GitHub Trending ·
★ 16.7k+75 today

Gives AI agents eyes to read and search Twitter, Reddit, YouTube and more via one CLI.

GitHub Trending ·
★ 89.8k+1.7k today

Makes your AI agent think like the laziest senior dev, avoiding unneeded code.

GitHub Trending ·
★ 153.4k+1.3k today

Agent harness performance optimization covering skills, memory, security for coding agents.

GitHub Trending ·
★ 272.2k+897 today

Small, composable agent skills for real engineering that work with any model.

GitHub Trending ·
★ 275.4k+751 today

A design language that makes your AI harness better at design.

GitHub Trending ·
★ 75.3k+699 today

An agentic skills framework and software development methodology for coding agents.

GitHub Trending ·
★ 294.9k+577 today

Skill and proxy for coding agents that cuts 65% of tokens by talking like a caveman.

GitHub Trending ·
★ 109.5k+507 today

AI agent toolkit with unified LLM API, agent loop, TUI and coding agent CLI.

GitHub Trending ·
★ 112.2k+408 today

Context window optimization for AI coding agents, sandboxing tool output and persisting memory.

GitHub Trending ·
★ 25.2k+256 today

Production-grade engineering skills for AI coding agents.

GitHub Trending ·
★ 100.8k+252 today

Agentic coding tool in your terminal that understands your codebase and runs tasks.

GitHub Trending ·
★ 149.2k+128 today

Agent workspace on Cloudflare Workers for creating documents, building apps, running agents.

GitHub Trending ·
★ 10.6k+85 today

Persistent context across sessions for agents, compressing and injecting relevant memory.

GitHub Trending ·
★ 95.6k+79 today

Model Ranking · 24 h

2026-10-04 UTC
01Space Bunny Alphastealth · 6.73T tokens29.5%—
02DeepSeek V4.1 FlashDeepSeek · 3.42T tokens15.0%—
03MiMo-V2.6-FlashXiaomi · 1.32T tokens5.8%—
04GLM 5.3 FlashZ.ai · 1.19T tokens5.2%—
05Hy4 previewTencent · 967.6B tokens4.2%▲1
06Nemotron 3 Ultra (free)NVIDIA · 915.6B tokens4.0%▼1
07GPT-6 LunaOpenAI · 768.8B tokens3.4%▲1
08DeepSeek V4 Flash 0423DeepSeek · 748.9B tokens3.3%▼1
09Jev 1.13typesafe · 401.0B tokens1.8%▲1
10DeepSeek V4 Flash 0423DeepSeek · 379.4B tokens1.7%▲1

New Models

3
Ling 3.1 FlashinclusionAI · 262.1k ctx · 10/02$0.00 / $0.00
Apodex 1.1 Mini (free)Apodex · 262.1k ctx · 10/01$0.00 / $0.00
Pareto 26.10 Previewunbiased · 1.0M ctx · 10/01$0.80 / $3.20

Deals

24 h · 1 deals